[CI] fix `test_concat_and_cache_mla_rope_fused` (#32117)

Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>

tests/kernels/core/test_rotary_embedding_mla_cache_fused.py

@@ -13,7 +13,7 @@ from tests.kernels.allclose_default import get_default_atol, get_default_rtol
 from tests.kernels.utils import DEFAULT_OPCHECK_TEST_UTILS, opcheck
 from vllm import _custom_ops as ops
 from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
-from vllm.platforms import current_platform
+from vllm.utils.torch_utils import set_random_seed
 
 
 @pytest.mark.parametrize("dtype", [torch.half, torch.bfloat16, torch.float])
@@ -31,6 +31,7 @@ from vllm.platforms import current_platform
 )
 @torch.inference_mode()
 def test_concat_and_cache_mla_rope_fused(
+    default_vllm_config,
     dtype: torch.dtype,
     is_neox_style: bool,
     seq_len: int,
@@ -45,7 +46,7 @@ def test_concat_and_cache_mla_rope_fused(
     max_position: int = 8192,
     base: float = 10000,
 ) -> None:
-    current_platform.seed_everything(seed)
+    set_random_seed(seed)
     torch.set_default_device(device)
 
     rope = RotaryEmbedding(